static int freeze_fs(struct ploop_copy_handle *h)
{
int ret;
if (h->mntfd != -1) {
/* Sync fs */
ploop_dbg(4, "SYNCFS");
if (sys_syncfs(h->mntfd)) {
ploop_err(errno, "syncfs() failed");
ret = SYSEXIT_FSYNC;
goto err;
}
/* Flush journal and freeze fs (this also clears the fs dirty bit) */
ploop_dbg(4, "FIFREEZE");
ret = ioctl_device(h->mntfd, FIFREEZE, 0);
if (ret)
goto err;
h->fs_frozen = 1;
}
ploop_dbg(4, "IOC_SYNC");
ret = ioctl_device(h->devfd, PLOOP_IOC_SYNC, 0);
if (ret)
goto err;
return 0;
err:
ploop_copy_release(h);
return ret;
}
static int ploop_trim(const char *mount_point, __u64 minlen_b, __u64 cluster)
{
struct fstrim_range range = {};
int fd, ret = -1;
struct sigaction sa = {
.sa_handler = stop_trim_handler,
};
sigemptyset(&sa.sa_mask);
if (sigaction(SIGUSR1, &sa, NULL)) {
ploop_err(errno, "Can't set signal handler");
exit(1);
}
fd = open(mount_point, O_RDONLY);
if (fd < 0) {
ploop_err(errno, "Can't open mount point %s", mount_point);
return -1;
}
sys_syncfs(fd);
if (minlen_b < cluster)
minlen_b = cluster;
else
minlen_b = (minlen_b + cluster - 1) / cluster * cluster;
range.minlen = MAX(MAX_DISCARD_CLU * cluster, minlen_b);
for (; range.minlen >= minlen_b; range.minlen /= 2) {
ploop_log(1, "Call FITRIM, for minlen=%" PRIu64, (uint64_t)range.minlen);
/* range.len is reseted by FITRIM */
range.len = ULLONG_MAX;
ret = ioctl(fd, FITRIM, &range);
if (ret < 0) {
if (trim_stop)
ret = 0;
else
ploop_err(errno, "Can't trim file system");
break;
}
/* last iteration should go with range.minlen == minlen_b */
if (range.minlen != minlen_b && range.minlen / 2 < minlen_b)
range.minlen = minlen_b * 2;
}
close(fd);
return ret;
}
static int freeze_fs(struct ploop_copy_handle *h)
{
int ret;
if (h->mntfd != -1) {
/* Sync fs */
ploop_log(0, "Freezing fs...");
if (sys_syncfs(h->mntfd)) {
ploop_err(errno, "syncfs() failed");
return SYSEXIT_FSYNC;
}
/* Flush journal and freeze fs (this also clears the fs dirty bit) */
ploop_dbg(3, "FIFREEZE");
ret = ioctl_device(h->mntfd, FIFREEZE, 0);
if (ret)
return ret;
h->fs_frozen = 1;
}
return 0;
}
int ploop_copy_send(struct ploop_copy_send_param *arg)
{
struct delta idelta = { .fd = -1 };
int tracker_on = 0;
int fs_frozen = 0;
int devfd = -1;
int mntfd = -1;
int ret = 0;
char *send_from = NULL;
char *format = NULL;
void *iobuf[2] = {};
int blocksize;
__u64 cluster;
__u64 pos;
__u64 iterpos;
__u64 trackpos;
__u64 trackend;
__u64 xferred;
int iter;
struct ploop_track_extent e;
int i;
pthread_t send_th = 0;
struct send_data sd = {
.mutex = PTHREAD_MUTEX_INITIALIZER,
.cond = PTHREAD_COND_INITIALIZER,
.cond_sent = PTHREAD_COND_INITIALIZER,
};
if (!arg)
return SYSEXIT_PARAM;
sd.fd = arg->ofd;
sd.is_pipe = is_fd_pipe(arg->ofd);
if (sd.is_pipe < 0) {
ploop_err(0, "Invalid output fd %d: must be a file, "
"a pipe or a socket", arg->ofd);
return SYSEXIT_PARAM;
}
if (arg->feedback_fd >= 0 && is_fd_pipe(arg->feedback_fd) != 1) {
ploop_err(errno, "Invalid feedback fd %d: must be "
"a pipe or a socket", arg->feedback_fd);
return SYSEXIT_PARAM;
}
/* If data is to be send to stdout or stderr,
* we have to disable logging to appropriate fd.
*
* As currently there's no way to disable just stderr,
* so in this case we have to disable stdout as well.
*/
if (arg->ofd == STDOUT_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOSTDOUT);
else if (arg->ofd == STDERR_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOCONSOLE);
devfd = open(arg->device, O_RDONLY);
if (devfd < 0) {
ploop_err(errno, "Can't open device %s", arg->device);
ret = SYSEXIT_DEVICE;
goto done;
}
mntfd = open_mount_point(arg->device);
if (mntfd < 0) {
/* Error is printed by open_mount_point() */
ret = SYSEXIT_OPEN;
goto done;
}
ret = get_image_info(arg->device, &send_from, &format, &blocksize);
if (ret)
goto done;
cluster = S2B(blocksize);
ret = SYSEXIT_MALLOC;
for (i = 0; i < 2; i++)
if (p_memalign(&iobuf[i], 4096, cluster))
goto done;
ret = complete_running_operation(NULL, arg->device);
if (ret)
goto done;
ret = ioctl_device(devfd, PLOOP_IOC_TRACK_INIT, &e);
if (ret)
goto done;
tracker_on = 1;
if (open_delta_simple(&idelta, send_from, O_RDONLY|O_DIRECT, OD_NOFLAGS)) {
ret = SYSEXIT_OPEN;
goto done;
}
ret = pthread_create(&send_th, NULL, send_thread, &sd);
if (ret) {
ploop_err(ret, "Can't create send thread");
ret = SYSEXIT_SYS;
goto done;
}
ploop_log(-1, "Sending %s", send_from);
trackend = e.end;
for (pos = 0; pos < trackend; ) {
int n;
trackpos = pos + cluster;
ret = ioctl_device(devfd, PLOOP_IOC_TRACK_SETPOS, &trackpos);
if (ret)
goto done;
n = do_pread(cluster, pos);
if (n == 0) /* EOF */
break;
async_send(n, pos);
pos += n;
}
/* First copy done */
iter = 1;
iterpos = 0;
xferred = 0;
for (;;) {
int err;
err = ioctl(devfd, PLOOP_IOC_TRACK_READ, &e);
if (err == 0) {
//fprintf(stderr, "TRACK %llu-%llu\n", e.start, e.end); fflush(stdout);
if (e.end > trackend)
trackend = e.end;
if (e.start < iterpos)
iter++;
iterpos = e.end;
xferred += e.end - e.start;
for (pos = e.start; pos < e.end; ) {
int n;
int copy = e.end - pos;
if (copy > cluster)
copy = cluster;
if (pos + copy > trackpos) {
trackpos = pos + copy;
if (ioctl(devfd, PLOOP_IOC_TRACK_SETPOS, &trackpos)) {
ploop_err(errno, "PLOOP_IOC_TRACK_SETPOS");
ret = SYSEXIT_DEVIOC;
goto done;
}
}
n = do_pread(copy, pos);
if (n == 0) {
ploop_err(0, "Unexpected EOF");
ret = SYSEXIT_READ;
goto done;
}
async_send(n, pos);
pos += n;
}
} else {
if (errno == EAGAIN) /* no more dirty blocks */
break;
ploop_err(errno, "PLOOP_IOC_TRACK_READ");
ret = SYSEXIT_DEVIOC;
goto done;
}
if (iter > 10 || (iter > 1 && xferred > trackend))
break;
}
/* Live iterative transfers are done. Either we transferred
* everything or iterations did not converge. In any case
* now we must suspend VE disk activity. Now it is just
* call of an external program (something sort of
* "killall -9 writetest; sleep 1; umount /mnt2"), actual
* implementation must be intergrated to vzctl/vzmigrate
* and suspend VE with subsequent fsyncing FS.
*/
/* Send the sync command to receiving side. Since older ploop
* might be present on the other side, we need to not break the
* backward compatibility, so just send the first few (SYNC_MARK)
* bytes of delta file contents. New ploop_receive() interprets
* this as "sync me" command, while the old one just writes those
* bytes which is useless but harmless.
*/
if (sd.is_pipe) {
char buf[LEN_STATUS + 1] = {};
ret = do_pread(4096, 0);
if (ret < SYNC_MARK) {
ploop_err(errno, "Short read");
ret = SYSEXIT_READ;
goto done;
}
TS("SEND 0 %d (sync)", SYNC_MARK);
async_send(SYNC_MARK, 0);
/* Now we should wait for the other side to finish syncing
* before freezing the container, to optimize CT frozen time.
*/
if (arg->feedback_fd < 0) {
/* No descriptor to receive a response back is given.
* As ugly as it looks, let's just sleep for some time
* hoping the other side will finish sync.
*/
TS("SLEEP 5");
sleep(5);
goto sync_done;
}
/* Wait for feedback from the receiving side */
/* FIXME: use select/poll with a timeout */
if (read(arg->feedback_fd, buf, LEN_STATUS)
!= LEN_STATUS) {
ploop_err(errno, "Can't read feedback");
ret = SYSEXIT_PROTOCOL;
goto done;
}
if (strncmp(buf, STATUS_OK, LEN_STATUS) == 0) {
goto sync_done;
}
else if (strncmp(buf, STATUS_FAIL, LEN_STATUS) == 0) {
ploop_err(0, "Remote side reported sync failure");
ret = SYSEXIT_FSYNC;
goto done;
}
else {
ploop_err(0, "Got back feedback: %s", buf);
ret = SYSEXIT_PROTOCOL;
goto done;
}
} else {
/* Writing to local file */
fdatasync(arg->ofd);
}
sync_done:
/* Freeze the container */
TS("FLUSH");
ret = run_cmd(arg->flush_cmd);
if (ret)
goto done;
/* Sync fs */
TS("SYNCFS");
if (sys_syncfs(mntfd)) {
ploop_err(errno, "syncfs() failed");
ret = SYSEXIT_FSYNC;
goto done;
}
/* Flush journal and freeze fs (this also clears the fs dirty bit) */
TS("FIFREEZE");
ret = ioctl_device(mntfd, FIFREEZE, 0);
if (ret)
goto done;
fs_frozen = 1;
TS("IOC_SYNC");
ret = ioctl_device(devfd, PLOOP_IOC_SYNC, 0);
if (ret)
goto done;
iter = 1;
iterpos = 0;
for (;;) {
int err;
struct ploop_track_extent e;
err = ioctl(devfd, PLOOP_IOC_TRACK_READ, &e);
if (err == 0) {
__u64 pos;
//fprintf(stderr, "TRACK %llu-%llu\n", e.start, e.end); fflush(stdout);
if (e.end > trackend)
trackend = e.end;
if (e.start < iterpos)
iter++;
iterpos = e.end;
for (pos = e.start; pos < e.end; ) {
int n;
int copy = e.end - pos;
if (copy > cluster)
copy = cluster;
if (pos + copy > trackpos) {
trackpos = pos + copy;
ret = ioctl(devfd, PLOOP_IOC_TRACK_SETPOS, &trackpos);
if (ret)
goto done;
}
TS("READ %llu %d", pos, copy);
n = do_pread(copy, pos);
if (n == 0) {
ploop_err(0, "Unexpected EOF");
ret = SYSEXIT_READ;
goto done;
}
TS("SEND %llu %d", pos, n);
async_send(n, pos);
pos += n;
}
} else {
if (errno == EAGAIN)
break;
ploop_err(errno, "PLOOP_IOC_TRACK_READ");
ret = SYSEXIT_DEVIOC;
goto done;
}
if (iter > 2) {
ploop_err(0, "Too many iterations on frozen FS, aborting");
ret = SYSEXIT_LOOP;
goto done;
}
}
/* Must clear dirty flag on ploop1 image. */
if (strcmp(format, "ploop1") == 0) {
int n;
struct ploop_pvd_header *vh;
TS("READ 0 4096");
n = do_pread(4096, 0);
if (n < SECTOR_SIZE) {
ploop_err(errno, "Short read");
ret = SYSEXIT_READ;
goto done;
}
vh = iobuf[i];
vh->m_DiskInUse = 0;
TS("SEND 0 %d (1st sector)", SECTOR_SIZE);
async_send(SECTOR_SIZE, 0);
}
TS("IOCTL TRACK_STOP");
ret = ioctl(devfd, PLOOP_IOC_TRACK_STOP, 0);
if (ret)
goto done;
tracker_on = 0;
TS("SEND 0 0 (close)");
async_send(0, 0);
pthread_join(send_th, NULL);
send_th = 0;
done:
if (send_th)
pthread_cancel(send_th);
if (fs_frozen)
(void)ioctl_device(mntfd, FITHAW, 0);
if (tracker_on)
(void)ioctl_device(devfd, PLOOP_IOC_TRACK_ABORT, 0);
free(iobuf[0]);
free(iobuf[1]);
if (devfd >=0)
close(devfd);
if (mntfd >=0)
close(mntfd);
free(send_from);
if (idelta.fd >= 0)
close_delta(&idelta);
TS("DONE");
return ret;
}
#undef do_pread
#undef async_send
/* Deprecated, please use ploop_copy_send() instead */
int ploop_send(const char *device, int ofd, const char *flush_cmd,
int is_pipe)
{
struct ploop_copy_send_param s = {
.device = device,
.ofd = ofd,
.flush_cmd = flush_cmd,
};
return ploop_copy_send(&s);
}